Unsupervised Behaviour Profiling

摘要: Given a large quantity of unprocessed videos object activities, the goal automatic behaviour profiling is to learn model that capable detecting unseen abnormal patterns whilst recognising novel instances expected normal patterns. In this context, an anomaly defined as atypical pattern not represented by sufficient examples in previous observations. Behaviour unsupervised learning and detection treated binary classification problem. One main challenges for differentiate true from outliers give false positives. chapter, we consider clustering discovers intrinsic grouping The method does require manual data labelling either feature extraction or discovery grouping. This crucial because often impractical given vast amount video data, subject inconsistency error prone. performs incremental cope with changes behavioural context. It also detects anomalies on-line so (a decision on whether made soon visual evidence collected without completion observed pattern.